Rafal Z bikowski
he effort to design and build a micro air vehicle flight inside buildings, stairwells, shafts, caves, and tunnels
T (MAV) with insect-like flapping wings has led to
a surprising new development in flight control,
which will accord miniature sensors a central
role. Unlike conventional aeronautics, where flight control is
achieved with fewer than 20 instruments, the new paradigm
of sensor-rich feedback control calls for hundreds of sensors.
The sensors will be networked to form numerous feedback
loops and to deliver highly informative signals. The signals
will require almost no further processing, so that real-time
flight control will involve very little computation. It is likely
that this is what insects do.
Micro Air Vehicles:
Indoor Reconnaissance
The concept of a hand-held (about 6 in or 15
cm) flying vehicle originated in 1996 at the
Defense Advanced Research Projects
Agency (DARPA). My research on
hand-held, autonomous flying vehi-
cles is motivated by a need for
intelligent reconnaissance
COREL
robots capable of discreetly
penetrating confined spaces
and maneuvering in them
without the assistance of a
human telepilot. Agile
September 2004 IEEE Instrumentation & Measurement Magazine
1094-6969/04/$20.002004IEEE
is of significant military and civilian value. The current
surveillance assets, such as satellites, manned aircraft, and
unmanned air vehicles, possess virtually no capability to
gather information inside buildings and other confined
spaces. Such capabilities will be particularly useful for
urban warfare, offering light-weight (less than 100 grams)
reconnaissance robots for individual soldiers. Tactical,
short-range surveillance is especially important for infantry
and special forces. Other applications include bomb dispos-
al, antiterrorist operations, and nuclear and chemical recon-
naissance, but also all other dull, dirty, or dangerous (DDD)
environments, where direct or remote human assistance is
not feasible, e.g., contaminated areas, nuclear reactors, and
machine rooms on ships. Nonmilitary uses of autonomous
MAVs will include law enforcement and rescue services,
such as people trapped under rubble. The human-free
exploration of industrial DDD environments will allow air
quality sampling in nonattainment areas, utility inspection
of pipes, and examination of confined spaces in buildings,
installations, and large machines.
Insect-Like Flapping Flight
The focus on indoor flight leads to the requirements of low
speed flight, agility at low speeds, minimal acoustic signa-
ture, vertical takeoff and landing, and autonomous flight, all
to be achieved in a power efficient way on a 6 in (15 cm)
19

(a)
STEPHEN DALTON/NHPA.
(b)
Fig. 1. (a) Insect wing kinematics, (b) Insect maneuvrability: a green lacewing
fly executing an upside-down vertical take off from a leaf; no man-made
flying machine is capable of such performance, especially at low speeds.
scale. As discussed elsewhere, the best-proven solution is
insect flight, which has been present in nature for over 300
million years [1][3]. The task is to implement the functional-
ity of insect flight by engineering means, as opposed to con-
structing an artificial flying insect.
Insects fly by oscillating (plunging) and rotating (pitch-
ing) their wings through large angles, while sweeping them
forwards and backwards. The wingbeat cycle, typically in
the frequency range 5200 Hz, can be divided into two phas-
es: downstroke and upstroke [Figure 1(a)]. At the beginning
of downstroke, the wing, as seen from the front of the insect,
is in the uppermost and rearmost position with the leading
edge pointing forward. The wing is then pushed down-
wards (plunged) and forwards (swept) and rotated (pitched)
continuously with considerable change of the angle of
20 IEEE Instrumentation & Measurement Magazine
attack. At the end of the downstroke, the wing is twisted
rapidly, so that the leading edge points backwards, and the
upstroke begins. During the upstroke, the wing is pushed
upwards and backwards and rotated again, which changes
the angle of attack throughout this motion. At the highest
point, the wing is twisted again, so that the leading edge
points forward, and the next downstroke begins.
The complex kinematics of flapping provide lift and
thrust, and allow remarkable maneuverability without mov-
able control surfaces [Figure 1(b)] [4]. Flies, in particular,
demonstrate impressive performance with only one pair of
wings, which is easier to emulate by engineering means than
two pairs. The wings are very light, typically 1% of the
insects weight, because, unlike birds, they carry no muscles.
Hence, there is no need for actuators on the wings in an
engineering implementation. Insect wings are made of spars
spanned by a thin membrane, and the spars protrude into
the insects body, where they are attached to muscles. The
wings deform in flight, and these deformations contribute to
the flys agility, but the main factor in its maneuvrability is
changes in wingbeat kinematics. True aerobatics [as seen in
Figure 1(b)] are achieved by subtly timed wingbeat asymme-
tries. The left and right wings always beat at the same fre-
quency, but may have different amplitudes (excursions of
the wing within a stroke) and reverse the direction of motion
at different phases of the beat cycle.
The aerodynamics involved in insect flight are very com-
plex, and the mathematical description is further complicat-
ed by significant wing deformation, called high-amplitude
aeroelasticity, and the segmented nature of the insect body,
which is a multibody problem [2]. The state-of-the-art aero-
nautical approach models such phenomena by coupled, non-
linear, partial differential equations whose solution requires
supercomputers calculating for days. On the other hand,
flight control commands probably originate from the central
complex of the flys brain, which has 3,000 nerve cells [5].
Nerve cells, or neurons, can be thought of as the smallest
computational units. Three thousand neurons, each inter-
preted as an on-off (binary) transistor, give no more compu-
tational power than possessed by a toaster. And yet, insects
are more agile than our aircraft equipped with superfast dig-
ital electronics. This realization made me interested in
reverse engineering of insect flight control.
Flight Control: Old and New
How do we go about controlling the flight of a modern
fighter aircraft? The answer is illustrated in Figure 2,
which shows that even for modern aircraft, with complex
flight dynamics, only a few sensors are used. Inevitably,
the mathematical expression of the complexity involved is
quite intricate. The resulting equations must be solved in
real time, which requires powerful and fast computers.
Although the equations are known, they do not necessari-
ly give much insight into how to achieve high perfor-
mance while ensuring flight stability and robustness
against disturbances and/or faults. Thus, not only is com-
September 2004
putation expensive, but, above
all, the controller design lacks
transparency.
There is no escaping the fact Commands Errors
that, both for the F-22 and a fly,
the flight dynamics are very
complex and thus require gener-
ating much information to
achieve effective flight control.
Hence, the first immutable con-
straint of the problem is the fact
that: complex flight dynamics
Fig. 2. Conventional flight control uses a few sensors and feedback loops and requires much computation.
require much information.
(F-22 courtesy of U.S. Air Force.)
Therefore the current para-
digm:
Conventional control:
a few sensors a few feedback loops a few
complex controllers
cannot be changed by simplifying the problemthe complex
dynamics are what makes high performance possible. Thus,
any alternative paradigm must generate the same amount of
information.
My hypothesis is that insects obey this constraint, but
they generate the required information by favoring measure-
ment rather than computation:
Sensor rich feedback control:
many sensors many feedback loops many
simple controllers
Hence, the paradigms may be contrasted as follows:
Conventional controlComplex flight dynamics: little
measurement/feedback data involved calculations
Sensor rich feedback controlComplex flight dynamics:
much measurement and feedback data simple cal-
culations
What biological evidence is there to support the sensor-
rich feedback control hypothesis? What measurable informa-
tion can be equivalent to the nonlinear differential equations
of flight dynamics?
Insect: A Sensor-Rich System
The fly brain receives sensory inputs from 80,000 receptor
axons and has 338,000 neurons [5]. The sensor-rich character
of this architecture is reflected by the fact that 98% of the
neurons are used for sensory processing, as opposed to gen-
eral-purpose computation, as in digital microprocessors. On
the other hand, there are only about a dozen wing muscles,
so the system is not actuator rich. Finally, the high-level inte-
gration of the sensory inputs happens in the small core of the
brain, the central complex, which has only 3,000 neurons.
Flight control of maneuvers is one of the high-level func-
tions, so 3,000 neurons is an upper bound on the computa-
tional resources available for such steering in the air.
September 2004 IEEE Instrumentation & Measurement Magazine
Inputs Wind
Accelerations
Controllers Altitude
20 Feedback Loops
The fly has many important tasks other than flying, so
there are many sensors not related to flight control, e.g., taste
and smell, touch, sound, and temperature and humidity. For
flying, the visual system and some mechanical sensors are of
critical importance. Most of the neural processing is devoted
to vision, and I shall focus on that later.
Before analyzing the sensor-rich aspects of insect vision,
it is worth mentioning that two-winged flies, Diptera, also
use a variety of mechanoreceptors for flight control [6].
Antennae and wind-sensitive hairs are obvious examples,
but, unlike other flying insects, Diptera have special sensors
on the thorax called halteres [7][9]. These are transformed
hind wings consisting of an end knob, a thin stiff stalk, and
an innervated socket in which they are mounted [Figure
3(c)]; halteres are sense organs for rotation. The ultra light
wings of the flies have no muscles (actuators), but do have
the so-called campaniform, or bell-shaped, sensilla on the
spars supporting the wing membrane; the sensilla measure
stresses [10][15]. Finally, the head is hinged to the thorax,
allowing considerable motions of the head relative to the
body; their relative position is sensed by the neck
mechanoreceptors [16], [17].
Insect Eyes:
Global Optic Flow Representation
Each of the flys compound eyes is composed of 8006,000
ommatidia, depending on the species [18]. Each ommatidi-
um is a miniature eye which measures light intensities with-
in a small solid angle (12 ) (Figure 4). This spatial
resolution is much lower than that of the human eye, but the
temporal resolution of photoreceptors in flies is higher by an
order of magnitude. Unlike pixel-base imaging cameras, the
compound eyes facilitate sophisticated representation of the
relative motion of the insect with respect to its surroundings.
An important fact is that the compound eyes allow sur-
veying practically the whole of the surrounding space, i.e.
the full 4 steradians of the sphere on which the space is
projected (Figure 4) [19]. Further, the ommatidia outputs are
processed locally by elementary motion detectors (EMDs).
The EMD signals are then integrated by tangential neurons
21

(a)
(b)
(c)
Fig. 3. Flight-related sensors of the housefly Musca domestica: a) the large
compound eyes dominate the head, but antennae and wind-sensitive hairs are
also present; b) ocelli are three light-sensitive sensors on top of the head;
c) halteres (specific only to two-winged flies) appear on the thorax and beat in
anti-phase to the wings, sensing rotary motion. (Images courtesy of Louis
DeVos, www.ulb.ac.be/science/biodic.)
to form a global vector field representing the relative motion
of the insect with respect to its surroundings [20]. This inte-
gration is done by at least 13 tangential neurons, and it is
22 IEEE Instrumentation & Measurement Magazine
Yaw Thrust
Slip
Roll Pitch
Lift
Fig. 4. The flys compound eyes allow the insect completely to survey the
surroundings, i.e. the full 4 steradians of the sphere on which the space is
projected [19] (Image courtesy of H.G. Krapp.).
remarkable that each tangential neuron represents half,
2 steradians, of the global vector field [19]; see Figure 5.
Each tangential neuron responds to all kinds of optic flows,
but is most sensitive to the flow corresponding to a specific
(preferred) direction of the insects motion (Figure 6).
Flight Dynamics from Overlapping
Representations of Global Optic Flow
Since the insect is more or less a rigid body flying in the air,
it is possible to represent its flight dynamics via six degrees
of freedom (6DOF) equations of motion, three equations for
the translational and three for rotational components [21].
This would be an external model, or flight dynamics as
observed from outside the insect. The first problem of exter-
nal modeling is to quantify the instantaneous aerodynamic
forces and moments present plus aeroelastic effects, while
the second is how to manipulate the forces and moments for
control. Insects do not integrate numerically such equations
in real time, but whatever they do must be equivalent to
having solutions of such equations. This would be an inter-
nal model or flight dynamics as seen by the insect. I believe
that the sensor-rich feedback control framework will allow
constructing an internal model and understand its relation to
the external description. If this is achieved, new vistas will
open in flight dynamics and in control in general. Since the
new paradigm relies on rich sensor input, this approach will
have a major impact on sensor and instrumentation applica-
tions and development.
So, how can the insect handle complex dynamics with-
out integrating differential equations? The answer is that it
already knows the solutions. An ordinary differential equa-
tion (ODE) quantifies how the rate of change (d/dt)(
x) = x

September 2004
of the variable of interest varies
when this variable and time t
d
do, i.e., x
= v
(
x, t) . In other d
75

Elevation
words, the ODE assigns to each At
vector x
and time t the unique
vector of the rate of change x
.
c 0 f c
This assignment is defined by f c
the right-hand side of the equa-
tion, i.e. v
= v
(
x, t). Such ODE 75
can be globally represented by 180 90 0 90 180
the defining vector field v
v Azimuth
v
= v
(
x, t), and the correspond-
(a)
ing solutions are the curves to
which the vectors of the field d
are tangent [22]. Thus, if the
75

Elevation
vectors are known with suffi- d
cient density in the (
x, t) c
domain, the solutions are readi-
c 0 f c
ly available. Hence the global
view of the optic flow (Figure 5) Ar
potentially provides the fly with 75
f
the full flight envelope of its v 180 90 0 90 180
dynamics relative to the sur-
v Azimuth
roundings. However, the way
this is done by insects seems to (b)
be quite subtle.
First, the vector field f
of the Fig. 5. Global representation of the vector field of the relative motion of the fly with respect to its surroundings,
optic flow is an encoded repre- encoded with optic flow: a) the fly translating upwards; b) the fly rotating along the long axis of its body. Notation:
sentation the vector field v
of ffrontal (where the head is), ccaudal (back), ddorsal (upper), vventral (lower), At axis of translation, Ar axis
the dynamics of relative motion of rotation [20]. (Image courtesy of H.G. Krapp.)
of the insect with respect to its
patterned surroundings. The
vector field f
of the optic flow (Figure 5) is a depiction of Second, the vector fields considered are formed on a
the kinematics of the relative motion. What must be inferred sphere, resulting in a second-order ODE on a manifold, a sit-
from these optic flow vectors is the dynamics of this motion, uation for which an extensive theory is available [22], [23]. A
i.e. the vector field v
= v
(
x, t) defining the equation of the manifold is a mathematical formalization of the notion of
dynamics x
= v
(
x, t). It does not necessarily follow that a smooth surface. A convenient way of thinking about a mani-
given optic flow vector field f
corresponds directly to the fold is by considering a geographical globe and a corre-
vector field v
, as this depends on the actual encoding per- sponding atlas. Each map is a straightened out patch of
formed by the tangential neurons. This encoding, or trans- the curved Earth. Such straightening out can be done unam-
formation between the vector fields f
and v
, is not well biguously if the surface has a unique tangent plane every-
understood, but hints about its nature are illustrated in where (smoothness). The globe represents the Earth as one
Figure 5. The orientations of vectors in Figure 5(a) are the curved surface, while the atlas needs at least two flat maps
directions of motion of the surrounding pattern, projected to accomplish that. If one flat map is used, as in the Mercator
on a sphere, in response to the insects upward translation. projection in Figure 5 (on the right), two antipodal points on
The sensitivities of the neural response are related to the the sphere are represented discontinuously: they become
lengths of these vectors. Similarly, the optic flow vector lines. To represent the whole world in a continuous and
field in Figure 5(b) corresponds to the motion of the sur- unambiguous way, the maps must overlap, so at least two
rounding pattern when the insect is in pure rotation. This is are needed. Thus a smooth surface can be represented by a
useful, as the 6DOF differential equation of the dynamics of collection of overlapping, rectifiable patches, and the sur-
relative motion x
= v
(
x, t) expresses the kinematic conse- faces global structure can be reconstructed from this collec-
quences of the aerodynamic forces and moments acting tion owing to the overlapping.
along three axes of translation and three axes of rotation. Among smooth manifolds, the two-dimensional sphere
The action of the forces and moments is mediated by the has particularly convenient global (topological) properties,
insects mass and inertia, which, presumably, are available as it is closed and bounded (compact, unlike the plane) and
to the insect. has no holes (has genus 0, unlike the torus). A remarkable

September 2004 IEEE Instrumentation & Measurement Magazine 23

consequence of this neat topological structure is that every
smooth vector field defined on the two-dimensional sphere
must have at least one point at which the corresponding
vector vanishes. This is the hairy ball theorem: there must
be a bald spot even on a ball with no hair missing, provid-
ed the combing is smooth [24]. Indeed, two such symmetri-
cal, singular points can be readily seen in Figure 5, both for
pure translation (a) and pure rotation (b). In fact, these are
elementary manifestations of the well-developed theory of
the singularity index of vector fields, which is particularly
powerful for the two-dimensional sphere [25][29]. From
the practical point of view, the insect needs only to detect
the point at which the encoded vector is zero, and this
alone will allow it to infer several global properties of its
flight dynamics.
Front
Left Right
VS2
VS1
VS10
VS3
VS9
VS8 VS4
VS7
VS6 VS5
Rear
Fig. 6. Preferred rotation axes of VS tangential neurons [19], i.e., directions
of insects motion generating the optic flow to which a given VS neurons
response is best matched. (Image courtesy of H.G. Krapp.)
Third, the insect can simultaneously measure at least 13
patches of the global vector field, each patch covering at
least a half-sphere. Recall that there are at least 13 tangen-
tial neurons involved, and each of them represents one
half (2 steradians) of the global vector field [19]. This
enables forming a system of several second-order equa-
tions, so that recovering a mere 6DOF can be readily
accomplished. There are, however, other roles for this
impressive redundancy. Not only does each tangential
neuron represent a half of the global vector field, but also
it does that with sensitivity dependent on the direction of
the insects motion (Figure 6). In other words, tangential
neurons are matched filters; they always respond to the
24 IEEE Instrumentation & Measurement Magazine
insects motion, but respond best (are tuned to) rotations
or translations along certain directions in space. The first
obvious role of these 13 matched filters is robustness to
uncertain or unavailable data, even when parts of the
vision system are damaged. More subtly, the absolute
dynamics of the fly in the three-dimensional Cartesian
space is represented by the insects relative motion with
respect to the surroundings, projected on a two-dimen-
sional sphere. The geometric ambiguities of the projection
will always be present, and additional ambiguities will
arise if parts of the environment are also in motion.
Resolution of these ambiguities is possible by simultane-
ous consideration of several overlapping patches of the
same vector field. This is aided by different scaling of the
vector lengths in the patches owing to the matched filter
nature of the tangential neurons. Moreover, the extreme
agility routinely demonstrated by two-winged flies is
impossible without sophisticated handling of inertial cou-
pling during maneuvers. This requires global and finely
grained feedback loops reacting to the interactions of
pitch, roll, and yaw. Spatially sensitive tangential neurons
may give the required information about the evolution of
the global vector field during the maneuvers. Several over-
lapping patches of the same vector field v
= v
(
x, t) also
allow the corresponding solutions of x
= v
(
x, t) to be inter-
polated more precisely, despite the time-varying character
of the problem. Another possibility is that the vector field
can be represented, not only by orthogonal decomposi-
tions, but also by more advanced schemes like the Hodge
decomposition [30], [31].
Finally, all these redundant and densely spaced represen-
tations of the vector field of the insect flight dynamics open
considerable possibilities for feedback control without the
need for heavy computation. Since several overlapping
patches (visual fields) are available simultaneously, a series
of small and thus easily controlled adjustments are needed
for the desired patterns of the vector field to be achieved.
It is worth noting that the aforementioned scheme will
not work if the environment is perfectly homogeneous, for
no meaningful relative motion information can be generated
then. Also, even in structured surroundings, there is a need
for up/down orientation, which insects seem to achieve with
ocelli (Figure 4). Mechanoreceptors are also needed for
implementing the control loops (e.g., the stress sensors on
wings) and possibly for guidance (e.g., neck sensors to orient
head and the rest of the body). Halteres are present only in
two-winged flies, so they are not absolutely necessary for
successful flight control, but two-winged flies are the best
flyers, so it is a performance-enhancing sensor. Thus, a full
sensor-rich feedback control architecture involves multisen-
sor data fusion. However, the key element is that this fusion
happens with very little computation; the architecture pro-
cesses the information to such a high degree that almost no
further calculations are needed. Effectively, all the important
quantities are produced by sensor systems according to the
idea if its difficult to compute, measure it.
September 2004
Summary
The available knowledge of insect flight seems to suggest
that two-winged flies implement a feedback control
paradigm not exploited in engineering and have remark-
able performance, achieved in a computationally simple
way. This sensor-rich feedback control paradigm relies on
extensive, distributed measurement of the quantities of
interest in space and time. The visual system of the fly rep-
resents its motion relative to the environment through
coarse, but still dense, representation of the vector field
involved instead of detailed imaging by fine pixels. This
representation is global, covers the whole of space, and
involves several overlapping patches. The lengths of the
vectors in the patches depend on the insects direction of
motion. Hence, the complex underlying differential equa-
tions of motion need not be integrated numerically, as
their solutions are readily available through interpolation
of the detailed, global representation of the relevant vector
field. Also, this redundant representation may allow fine
control of high agility maneuvers, despite inertial cou-
plings, with little computation. Indeed, a series of small
adjustments should suffice for the desired vector field pat-
terns to be achieved.
From the engineering viewpoint, this opens not only new
perspectives in control, but also in sensors and instrumenta-
tion. It suggests that there is no need for high-resolution
imaging cameras for flight control based on optic flow.
Instead, coarse-grained arrays of sensors could give good
results provided they are
arranged to offer a global view of the environment
endowed with parallel processing for extraction of the
global vector field of motion via several, overlapping
patches of the field
characterized by sensitivities dependent on the direc-
tion of motion.
The use of other sensors is likely to be needed, and this will
call for multisensor data fusion. However, the sensor-rich
feedback control architecture works on the complementarity
of the type of information measured, not the type of instru-
ment used. A gravity sensor or an altimeter, for example, can
replace ocelli.
So, the issue is not to emulate insect sensors, but to use
appropriate instrumentation to extract the relevant informa-
tion with the required speed and accuracy. Further, detailed
and multidisciplinary research is needed to reverse engineer
insect flight control completely. However, what we know
already hints of exciting possibilities of understanding
motion control in animals and humans and acquiring a new
paradigm in control engineering. The latter will have a large
impact not only on the vast applications of automatic con-
trol, but also on the sensor, instrumentation, and measure-
ment communities.
Acknowledgments
I thank Holger Krapp of Cambridge University and Graham
Taylor of Oxford University for their helpful comments on
September 2004 IEEE Instrumentation & Measurement Magazine
the manuscript. I am grateful to Neal Glassman and Belinda
King from AFOSR and Johnny Evers from AFRL from their
support for this work.
References
.
[1] R. Zbikowski, Flapping wing technology, in Proc. European
Military Rotorcraft Symp., Shrivenham, U.K., March 2123, 2000,
pp. 17.
.
[2] R. Zbikowski, On aerodynamic modeling of an insect-like
flapping wing in hover for micro air vehicles, Philosoph. Trans.
Roy. Soc. London (Series A: Math., Physical and Eng. Scis.), vol. 360,
no. 1791, pp. 273290, 2002.
[3] R. Dudley, The Biomechanics of Insect Flight: Form, Function,
Evolution. Princeton, NJ: Princeton Univ. Press, 2000.
[4] S. Dalton, Caught in Motion. London, England, Weidenfeld &
Nicolson, 1982.
[5] N.J. Strausfeld, Atlas of an Insect Brain. Berlin and New York,
Springer-Verlag, 1976.
[6] T.A. Keil, Functional morphology of insect mechnoreceptors,
Microscopy Res. Technique, vol. 39, no. 6, pp. 506531, 1997.
[7] G. Nalbach, The halteres of the blowfly Calliphora. I. Kinematics
and dynamics, J. Comparative Physiology A, vol. 173, no. 3, pp.
293300, 1993.
[8] G. Nalbach and R. Hegstenberg, The halteres of the blowfly
Calliphora. II. Three-dimensional organization of compensatory
reactions to real and simulated rotations, J. Comparative
Physiology A, vol. 175, no. 6, pp. 695708, 1994.
[9] G. Nalbach, Extremely non-orthogonal axes in a sense organ for
rotation: Behavioural analysis of the Dipteran haltere system,
Neuroscience, vol. 61, no. 1, pp. 149163, 1994.
[10] M.H. Dickinson and J. Palka, Physiological properties, time of
development, and central projection are correlated in the wing
mechanoreceptors of Drosophila, J. Neuroscience, vol. 7, no. 12,
pp. 42014208, 1987.
[11] M.H. Dickinson, Linear and nonlinear encoding properties of
an identified mechanoreceptor of the fly wing measured with
mechanical noise stimuli, J. Experimental Biology, vol. 151, pp.
219244, 1990.
[12] M.H. Dickinson, Comparison of encoding properties of
campaniform sensilla of the fly wing, J. Experimental Biology,
vol. 151, pp. 245261, 1990.
[13] M.H. Dickinson, Directional sensitivity and mechanical
coupling dynamics of campaniform sensilla during chordwise
deformations of the fly wing, J. Experimental Biology, vol. 169,
pp. 221233, 1992.
[14] W. Gnatzy, U. Grnert, and M. Bender, Campaniform sensilla
of Calliphora vicina (Insecta, Diptera). I. Topography,
Zoomorphology, vol. 106, no. 5, pp. 312319, 1987.
[15] U. Grnert and W. Gnatzy, Campaniform sensilla of Calliphora
vicina (Insecta, Diptera). II. Typology, Zoomorphology, vol. 106,
no. 5, pp. 320328, 1987.
[16] R. Hegstenberg, Mechanosensory control of compensatory
head roll during flight in the blowfly Calliphora erythrocephala
Meig., J. Comparative Physiology A, vol. 163, no. 2, pp. 151165,
1988.
[17] R. Hegstenberg, Multisensory control in insect oculomotor
25
 system, in Visual Motion and Its Role in the Stabilization of Gaze, F.
A. Miles and J. Wallman, Eds. Amsterdam: Elsevier, 1993, pp.
285298.
[18] J.G.J. Smakman, J.H. van Hateren, and D.G. Stavenga, Angular
sensitivity of blowfly photoreceptors: Intracellular
measurements and wave-optical predictions, J. Comparative
Physiology A, vol. 155, no. 2, pp. 239247, 1984.
[19] H.G. Krapp, Neuronal matched filters for optic flow processing
in flying insects, in Neuronal Processing of Optic Flow,
(International Review of Neurobiology, vol. 44), M. Lappe, Ed.
San Diego: Academic, 2000, pp. 93120.
[20] H.G. Krapp and R. Hegstenberg, Estimation of self-motion by
optic flow processing in single visual interneurons, Nature, vol.
384, no. 6608, pp. 463466, 1986.
[21] G.K. Taylor and A.L.R. Thomas, Dynamic flight stability in the
desert locust Schistocerca gregaria, J. Experimental Biology, vol.
206, no. 16, pp. 28032829, 2003.
[22] V.I. Arnold, Ordinary Differential Equations. Berlin: Springer-
Verlag, 1992.
[23] C. Godbillon, Dynamical Systems on Surfaces. Berlin: Springer-
Verlag, 1983.
[24] M. Eisenberg and R. Guy, A proof of the hairy ball theorem,
Amer. Math. Monthly, vol. 86, no. 7, pp. 572574, 1979.
[25] D.H. Gottlieb, Vector fields and classical theorems of
topology, Rendiconti del Seminario Matematico e Fisico di Milano,
vol. 60, pp. 193203, 1990.
[26] K.R. Meyer, On computing the index in three dimensions,
Proc. Amer. Math. Soc., vol. 19, pp. 760763, 1968.
[27] A.J. Schwartz, A generalization of a Poincar-Bendixson
theorem to closed two-dimensional manifolds, Amer. J. Math.,
vol. 85, pp. 453458, 753 (errata), 1963.
[28] J. Milnor, The characteristics of a vector field on the two-
sphere, Ann. Math., vol. 58, no. 2, pp. 253257, 1953.
[29] J.F. Adams, Vector fields on spheres, Ann. Math., vol. 75, no. 3,
pp. 603632, 1962.
[30] A. Szustalewicz, On orthogonal decomposition of two-
dimensional vector fields, Applicationes Mathematicae, vol. 38,
no. 3, pp. 433472, 1984.
[31] J. Cantarella, D. DeTurck, and H. Gluck, Vector calculus and
the topology of domains in 3-space, Amer. Math. Monthly, vol.
109, no. 5, pp. 409442, 2002.
Rafal Zbikowski (R.W.Zbikowski@cranfield.ac.uk) is a prin-
cipal research officer in the Department of Aerospace, Power,
and Sensors at Cranfield University, Shrivenham campus, the
Royal Military College of Science, England. A control engi-
neer with a strong mathematical background, he initiated the
flapping wing micro air vehicles research in the U.K. and has
led the effort since early 1998. He is a Member of the IEEE,
the American Mathematical Society (AMS) and the American
Institute of Aeronautics and Astronautics (AIAA).

A
Technology and Tradition
Unite In San Antonio
20 - 23 September 2004
Henry B. Gonzales Convention Center

http://www.autotestcon.com

IEEE AUTOTESTCON is the worlds only conference that focuses primarily on Automated Test and related technology for
military, government and aerospace applications. Over its 35+ year existence it has grown into the premier annual conference
for this application field. The conference also has an expanded focus into commercial areas that share a common technical
base, including aerospace, vehicle and automotive, and commercial factory test applications.

AUTOTESTCON is a comprehensive technical conference that provides an open forum for leaders in design, development,
procurement, applications and operations to exchange information relative to their specific needs and disciplines concerning
automated and computer-controlled test systems and software. Heavy focus is given to manufacturing and maintenance
testing environments. In addition to the comprehensive Technical Program, there is a full suite of directly related exhibits that
provide significant enhancement to the technical focus.

AUTOTESTCON also hosts a Seminars Program that provides a number of tutorials in the Automatic Test field, covering
hardware, software, and Test Program Set topics of interest. Todays demanding test environments require the most up-to-
date hardware and software. Come to AUTOTESTCON in San Antonio, TX and see it all in one place!

26 IEEE Instrumentation & Measurement Magazine September 2004